Remote sensing of wetlands in the Lake Whangape catchment, Waikato, New Zealand.

Abstract

Wetlands are among the world's most valuable ecosystems. They provide numerous ecological and socio-economic benefits. However, wetlands continue to disappear due to the increasing demand for wetland resources. In New Zealand, more than 90% of the original extent of wetlands has been lost since the mid-eighteenth century. Therefore, legislation has been identified for the protection of wetlands as a matter of national importance.

Geographic Information System (GIS) and Remote Sensing (RS) techniques have proven helpful for mapping and monitoring wetland resources. This study aims to understand how RStechniques can classify wetlands in the Lake Whangape catchment, Waikato. The parameters that can be extracted from available data and their effectiveness in the classification process are also studied.

Four types of input data are collectively employed in the study. The data types are optical RS data, Synthetic Aperture Radar (SAR) data, a Digital Elevation Model (DEM), and wetland polygons provided by the Waikato Regional Council (WRC). All the steps including, accessing satellite scenes and data processing were performed within Google Earth Engine (GEE) computing platform using JavaScript language. The classification process for this study includes feature extraction, feature selection, model training, classification, and validation. Finally, the accuracy of the classification results is checked visually and statistically.

The classification was carried out in two stages. In Stage one, open water, wetland, and non-wetland areas are classified (simple classification). The combined wetlands class is separated into marsh and swamp in the second stage (detailed classification). Based on the results, the Topographic Position Index (TPI) is the most influential parameter in identifying wetlands, while the Modified Normalized Water Index (MNDWI) successfully identifies open water. The overall accuracy reached 91% at the simple classification stage. However, the detailed classification results received comparatively low classification accuracies (the overall accuracy is 76%).